<< All versions

Skill v1.0.1

currentAutomated scan100/100
scriptedalchemy/pstack-codex/automate-team

+3 new

──Details
PublishedOctober 22, 2026 at 03:50 PM
Content Hashsha256:3ed2ebb9077420a7...
Git SHA139e929c47ff
Bump Typepatch
Compare with v1.0.0
──Files
Files (1 file, 3.3 KB)
SKILL.md3.3 KBactive
SKILL.md · 38 lines · 3.3 KB

version: "1.0.1" name: automate-team description: "Discover a repository's active core contributors and create or refresh a team mode from their engineering activity. Use for automate team or capture this repository's maintainer conventions."


Before following this workflow, read the Codex runtime contract.

Automate team

Discover a repository's active core contributors and build a mode from their observed engineering conventions. Use automate-maintainer for mining, attribution, group synthesis, authoring, and evaluation; read it before collecting training evidence.

Establish scope

Accept a repository URL or owner/repo, otherwise resolve the checkout's Git remote. Ask which repository only when remotes make the intended scope unclear. Verify visibility and default branch through an authenticated GitHub connector or gh. Default to the last 12 months; several repositories need separate ownership and visibility records. Linked projects supply context without expanding the crawl.

Discover the roster

Read roster discovery. Compare declared ownership with substantive authorship, reviews, design discussions, triage, and release activity. Exclude bots and distinguish active owners from inactive declared maintainers and uncertain candidates. Counts alone do not establish a core role.

Report the inferred roster with ownership evidence and proceed with well-supported accounts. Honor supplied inclusions and exclusions. Ask only when identity, an uncertain inclusion, scope, or crawl size changes the work. Do not silently reduce “all core contributors” to the busiest few. Process large rosters in batches, and record every included person and any unfinished coverage.

Mine, draft, and validate

Pass the roster, repository, window, ownership evidence, destination, and omissions to automate-maintainer. Reserve held-out evaluation cases before reading their conversations. Collect each person's record separately; delegate only when allowed, with bounded scope and output ownership.

Default to one <repo>-team-mode with individual evidence in references. Create separately invocable individual modes when requested. Preserve an existing team mode's name, category, and invocation policy.

State shared conventions, supported ownership, and disagreements. Repository requirements can apply across the team without proving each contributor's preference. Do not claim unanimity from missing evidence or resolve conflicts by activity counts.

Validate through skill-creator and automate-maintainer's evaluation procedure. Report roster and source coverage, unsupported areas, and validation limits. Land within existing authorization; creating a mode does not launch a team, authorize communication as its members, or schedule crawling.

Refresh

Compare roster and ownership with the previous evidence cutoff. Reassess departed or inactive contributors, new owners, changed conventions, and counterexamples. Preserve supported rules; remove contradicted claims. Record membership changes and the new cutoff so the next refresh can continue without repeating the crawl.

Use automate-maintainer for a supplied person or group, and automate-me for the user's conversation history.

← v1.0.0All versions